Tag
An opinion article arguing that as AI commoditizes many skills, ethical judgment becomes the most important human skill, and that ethics is a trainable skill built from noticing, reasoning under conflict, owning second-order effects, and acting under pressure.
This paper introduces a psychometric battery to separate framing artifacts from genuine moral judgment in LLMs, finding that frontier models have a coherent internal moral scale but display a yes/no bias that is purely a surface-level artifact of answer order and wording, not a real disposition to reject.
This paper introduces a bidirectional diagnostic, Compliance Asymmetry, and finds that LLMs exhibit 'directional blindness' in moral judgments: they comply equally to helpful and harmful social nudges, unlike in factual domains where they selectively follow helpful corrections. The phenomenon persists across models and nudge types, highlighting a distinct failure mode in current LLM alignment.